01政策相关
102行业前沿动态
5Google DeepMind 发布 Gemini 3.8 Live 和 3.8 Live Extended Thinking
2026-09-15Google DeepMind 发布 Gemini 3.8 Live 与 Gemini 3.8 Live Extended Thinking 两个近实时语音对话模型,主打语音智能体和复杂任务执行。
Google 发布 TranslateGemma 等多语言 AI 成果,语言技术覆盖 300 多种语言
2026-09-15Google 宣布其语言技术已支持超过 300 种语言、覆盖全球 86% 人口,并发布 TranslateGemma 轻量开源翻译模型(基于 Gemini 训练、支持 55 种语言、可离线运行)。
生数科技发布 Vidu S2:含 Avatar 与 Editing 双模型,探索空间视频
2026-09-15生数科技正式发布 Vidu S2,包含面向数字角色实时交互的 Vidu S2-Avatar 和面向视频流实时编辑的 Vidu S2-Editing,并探索面向 VR 头显的实时空间视频生成与编辑。
阶跃星辰发布 StepAudio 3 系列语音大模型,多款在 Artificial Analysis 榜单全球第一
2026-09-15阶跃星辰发布 StepAudio 3 系列,包含 Realtime、ASR、TTS、Gen 和 Music 五款模型,已在阶跃星辰开放平台上线。
Fireworks 评测 DeepSeek-V4.1-Flash:DeepSWE 达 GPT-6 Astra 水准、成本仅 1/15
2026-09-15Fireworks 发布 DeepSeek-V4.1-Flash 并公布完整基准结果:在 DeepSWE 上以 max 档取得 74.34% pass@1,与 GPT-6 Astra 同一水平,但每任务成本 $0.43,约为 Astra 的 1/15。
04论文研究
8When Can You Trust Your Synthetic Users? Diagnostics and Corrections for LLM Consumer Panels
2026-09-15arXiv:2609.13148v1 Announce Type: new Abstract: Large language models are increasingly deployed as synthetic consumer panels, promising $97\%$ cost reductions over traditional surveys. Yet aggregate validation metrics co…
PAUSE: A Privacy-Preserving Self-Reflection Tool for AI-Associated Cognitive Offloading
2026-09-15arXiv:2609.13155v1 Announce Type: new Abstract: Cognitive offloading is the use of external aids, such as notes, calculators, or search engines, to reduce mental effort. Large language models (LLMs) extend this to thinki…
picoRing dual: ultralow-power bimanual ring controllers for everyday AR
2026-09-15arXiv:2609.13160v1 Announce Type: new Abstract: Advances in smart eyewear, such as AR glasses and HMDs, bring digital information into outdoor and public situations. However, traditional handheld controllers for hand int…
The Landscape of Misinformation Literacy Games: A Systematic Mapping Review of Game Designs and Evaluation Practices
2026-09-15arXiv:2609.13162v1 Announce Type: new Abstract: This systematic mapping review examines 54 gamified interventions aimed at combating misinformation. We analyze the theoretical frameworks, game design, and media contexts …
Yeah, I Can Play That - Aesthetic Education in Kindergarten Using a Social Robot
2026-09-15arXiv:2609.13164v1 Announce Type: new Abstract: There have been numerous attempts to investigate the potential of social robots for education in the field of child-robot interaction (cHRI). However, empirically based mod…
User-Side Contextual Phenomena in Long-Term Human-AI Interaction
2026-09-15arXiv:2609.13165v1 Announce Type: new Abstract: Current assessments of conversational AI focus mainly on model outputs, including hallucinations and factual errors. These measures matter, but this paper examines risks th…
Code in Space: How Multimodal Human-AI Experience Can Reshape the Future of Tech Creation
2026-09-15arXiv:2609.13167v1 Announce Type: new Abstract: Advances in artificial intelligence (AI) continue to reshape digital product development, yet the day-to-day tools for developers and designers remain bound to flat screens…
A Cross Community Agenda for Speech AI
2026-09-15arXiv:2609.13168v1 Announce Type: new Abstract: Speech AI, any AI system that recognizes, transforms, or generates speech, is built and evaluated across two communities with only a small overlap: technical natural langua…
05技巧与观点
7Arena 更新 Image-to-WebDev 榜单:GPT-6 Astra 以 1733 分登顶
2026-09-15Arena 更新 Image-to-WebDev 排行榜,纳入四个新评测模型。GPT-6 Astra (Max) 以 1733 分居第一,领先 GPT-5.6 Sol (xHigh) 129 分;Claude Fable 5.1 (Max) 以 1710 分排第二,Muse Spark 1.3 (Max) 1645 分第四,GLM-5.3-Flash 1588 分第十。
Gergely Orosz 探访 OpenAI:Codex 驱动的智能体软件工厂
2026-09-15Gergely Orosz 实地探访 OpenAI 总部并访谈七位工程师与工程负责人,发现自约一月起 Codex 和 ChatGPT Work 已成为公司几乎所有工作的基础。
404 Media 曝光 OpenAI 莉莉计划:人工审核 ChatGPT 聊天记录以优化模型
2026-09-15404 Media 披露 OpenAI 内部代号为莉莉计划(Project Lily)的项目,由时薪超 50 美元的提示词审核员查看匿名化后的真实用户聊天记录,评判回复是否切题、是否存在 AI 式话术和谄媚口吻。
Trail of Bits 批评 1Password 的 AI 补丁基准存在误导,并发布两个补丁验证 Agent 技能
2026-09-15Trail of Bits 发文批评 1Password 8月6日发布的 FLAWED 报告,称其 26% 的 AI 干净修复率受四项实验设计选择影响而失真,包括刻意指示智能体应用错误修复的提示词占 22% 数据、36% 的试验禁止编译测试,以及不同推理档位设置。
Anthropic 与 OpenAI 提议协调放缓前沿 AI 开发,Cohere CEO 等批评者质疑其真实动机
2026-09-15Anthropic CEO Dario Amodei 呼吁行业与政府协调放缓前沿 AI 开发,并寻求反垄断豁免,Sam Altman 与 Elon Musk 表示同意。
Artificial Analysis 评测:GPT-Live-1 以 81.5 分登顶 Speech to Speech Index
2026-09-15Artificial Analysis 发布 Speech to Speech Index,OpenAI 的 GPT-Live-1 以 81.5 分(Astra 后端,medium 推理强度)排名第一,超过 Grok Voice Think Fast 2.0 High 的 81.3;Sol 后端配置得 80.1 排第三。
Vercel 将 inbound 销售团队从 10 人压缩至 1.25 人,AI 销售开发智能体年成本仅数千美元
2026-09-15Vercel COO Jeanne DeWitt Grosser 在 The Information 访谈中表示,公司 inbound 销售开发已实现 90% 自动化,团队从 10 人压缩至 1.25 人。
